dga: Perhaps Wolf's figure of 1201 hps/c3.8xl was an outlier, then, if 1000 is typical. But your spot instance rate seems a bit high, since the amazon site persistently reports in the .256 to .288 range. Any light on that?
nakaone: I completely agree.
I like to track the distribution of mining costs in order to model how each typical demographic will behave. If you had the relative numbers of miners in each major bucket, and could estimate their sales timing, it would give you a substantial market timing edge. We're getting pretty close to that point here, but not quite. I could try to parameterize a model from the price data, but the signal is low enough and the dimension of the parameter space high enough so that current history length would be inadequate to provide statistically robust parameters. The more dimension can be reduced, the sooner a valid parameter set can be derived.
The EC2 prices are back down - I think the EC2 miners are getting edged out by the botnets and GPUs at this point. (Reference: There was a self-claimed botherder in #monero with about 500kh/s. Not verified, but his numbers were sound.)
This is the Amazon spot price history for the last month for c3.8xlarge. It's pretty clear when XMR mining was very profitable:
(The three colors are three different zones. If only one spikes, it probably doesn't mean the overall price is high, but when all of them go high, it means that there's huge demand on EC2)
And then you can see that about a week ago, the spot prices returned back to near the "floor" (Amazon sets a minimum price for each instance type, even if there's no demand). That correlated roughly with a large increase in diff and drop in price of XMR.
Wolf's miner is no longer the one to use as a comparison, btw -- yam is faster. I havent benchmarked it on EC2, though, because the margin is too small (or negative) for my tastes. 1050 is probably a reasonable ballpark.
